b'David Annetts best of Exploration GeophysicsFeaturethan a complex distribution of conductivity. For example, a tabular body requires only 10 parameters (3 for location, 3 for attitude, 3 for size and one conductivity). Finite-element or finite-difference models will tend to have a larger number of parameters, but frequently many of these parameters such as cell conductivity will be the same. Once overburden or conductive host is included, the response becomes more complicated and program execution times increase.The inductive and resistive limit responses of single plate like targets are relatively easy to fit (Stolz and Macnae 1998; King 1998). Figure 5 shows a test of this fitting approach on relatively complex synthetic data, using Program EMFlow to interactively fit a plate model to synthetic horizontal (x) and vertical (z) component fixed-wing AEM data. The raw time-domain data (not shown) were first deconvolved to tau domain and the resistive limit calculated. Each component in Figure 5 Figure 4.Decomposition of 15 channel AEM data (bottom) into the resistivecontains a double-peaked resistive limit anomaly which can limit (top), showing the contribution from short (early), middle and slow (late)not be fitted by a single plate-like source. Two plates, however, tau values. provide a virtually perfect fit to the calculated resistive limit and accurately recover the original locations of the plates used A key aspect to the application of these limits is the recognitionin forward modelling. The model parameters listed are for the that it is possible to estimate the inductive limit of a local targetsolid dipping target on the plot. The inversion process takes under, but not in contact with, conductive cover. Conductiveseveral seconds per iteration for two plate-like bodies, but cover not in contact with a target has two effects: it firstly delaysmanual interaction is required in practice to force the solution the propagation of primary EM fields through it to energiseaway from local minima by choosing a reasonable starting the target, and further delays the secondary field of inducedmodel. Such local minima typically provide very poor fits to target currents during their passage back to an EM receiver.the data. The manual interaction involves adding a second These delays and associated time-smoothing lead to theconductor if a poor fit is obtained with a single conductor, and second effect: amplitude attenuation in frequency domain ormaking a fairly obvious choice as to its starting location under in repetitive time domain waveforms. The estimation methodthat part of the anomaly not well fitted by a single conductor. we use involves fitting field decays with time-retarded andWith AEM systems having a footprint limited to about 300 m attenuated decays based on model data, and then using(Xie, 1998), we usually choose to fix the strike length of any the model data to predict the free-space inductive limit. Themodel to be of this order. This implies that the fitting is most method is described in detail in King (1998), and is conceptuallyapplicable to targets of strike length exceeding 200 to 300 m.the inverse of the approximate forward modelling method ofIt is possible to numerically calculate the inductive limit of Liu and Asten (1993). any body in free space, using potential-field algorithms that enforce the boundary condition that no magnetic field can Fast forward and inverse models penetrate into the conductor (King 1998). A forward model takes about 10 seconds on a fast PC. Figure 6 shows an example To be able to fit a response to every local anomaly on an AEMof fitting a wide dipping target to the inductive limit predicted survey, it may be necessary to model on the order of 500 localfrom x-component Questem data. About 5 to 10 minutes anomalies per AEM system per day (Macnae et al. 1998). Clearly,on a Pentium PC is typically required for a such a fit. In this if one computer is used to process the data from one AEMcase, the response has not yet converged but has exceeded a system, the modelling process is limited to more than a minute or two per anomaly if it is to keep up with data acquisition. This places a severe constraint on which models can be used to interpret the data.Rapid forward models that are presently available are based on both mathematical and physical convenience, e.g., an assumption of uniform conductivity, and on geological constraints such as an assumption of a typical shape. Massive sulphide deposits are commonly tabular due to their origin as sediments from hydrothermal activity at the sea floor, and are commonly represented by plate-like models. Programs Plate (Dyck, Bloore, and Vallee 1980) and MultiLoop (Lamontagnc, Macnae, and Poller 1988) are examples of rapid computer programs which execute forward models in seconds. Most 2D or 3D conductivity structure models for a 3D source however execute in hours rather than seconds and are too slow to consider for routine application.Figure 5.A fit of two plate-like bodies: about two minutes on a Pentium PC Many EM algorithms are based on parameterised models, andwas taken to fit this anticline model to resistive limit AEM synthetic data using are represented by a very small number of parameters ratherboth the x and z; components. Ihe profile is 2 km long.OCTOBER 2020 PREVIEW 48'